Datasets:
Add dataset loader script
Browse files- jam-alt.py +124 -0
jam-alt.py
ADDED
@@ -0,0 +1,124 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""HuggingFace loading script for the JamALT dataset."""
|
2 |
+
|
3 |
+
|
4 |
+
import csv
|
5 |
+
from dataclasses import dataclass
|
6 |
+
import json
|
7 |
+
import os
|
8 |
+
from pathlib import Path
|
9 |
+
from typing import Optional
|
10 |
+
|
11 |
+
import datasets
|
12 |
+
|
13 |
+
|
14 |
+
# TODO: Add BibTeX citation
|
15 |
+
_CITATION = """\
|
16 |
+
"""
|
17 |
+
|
18 |
+
# TODO: Add description of the dataset here
|
19 |
+
_DESCRIPTION = """\
|
20 |
+
"""
|
21 |
+
|
22 |
+
# TODO: Add a link to an official homepage for the dataset here
|
23 |
+
_HOMEPAGE = ""
|
24 |
+
|
25 |
+
# TODO: Add the licence for the dataset here
|
26 |
+
_LICENSE = ""
|
27 |
+
|
28 |
+
_METADATA_FILENAME = "metadata.csv"
|
29 |
+
|
30 |
+
|
31 |
+
_LANGUAGE_NAME_TO_CODE = {
|
32 |
+
"English": "en",
|
33 |
+
"French": "fr",
|
34 |
+
"German": "de",
|
35 |
+
"Spanish": "es",
|
36 |
+
}
|
37 |
+
|
38 |
+
|
39 |
+
@dataclass
|
40 |
+
class JamAltBuilderConfig(datasets.BuilderConfig):
|
41 |
+
language: Optional[str] = None
|
42 |
+
with_audio: bool = False
|
43 |
+
decode_audio: bool = True
|
44 |
+
sampling_rate: Optional[int] = None
|
45 |
+
mono: bool = True
|
46 |
+
|
47 |
+
|
48 |
+
# TODO: Name of the dataset usually matches the script name with CamelCase instead of snake_case
|
49 |
+
class JamAltDataset(datasets.GeneratorBasedBuilder):
|
50 |
+
"""TODO: Short description of my dataset."""
|
51 |
+
|
52 |
+
VERSION = datasets.Version("0.0.0")
|
53 |
+
BUILDER_CONFIG_CLASS = JamAltBuilderConfig
|
54 |
+
BUILDER_CONFIGS = [JamAltBuilderConfig("default")]
|
55 |
+
DEFAULT_CONFIG_NAME = "default"
|
56 |
+
|
57 |
+
def _info(self):
|
58 |
+
feat_dict = {
|
59 |
+
"name": datasets.Value("string"),
|
60 |
+
"text": datasets.Value("string"),
|
61 |
+
"language": datasets.Value("string"),
|
62 |
+
}
|
63 |
+
if self.config.with_audio:
|
64 |
+
feat_dict["audio"] = datasets.Audio(
|
65 |
+
decode=self.config.decode_audio,
|
66 |
+
sampling_rate=self.config.sampling_rate,
|
67 |
+
mono=self.config.mono,
|
68 |
+
)
|
69 |
+
|
70 |
+
return datasets.DatasetInfo(
|
71 |
+
description=_DESCRIPTION,
|
72 |
+
features=datasets.Features(feat_dict),
|
73 |
+
supervised_keys=("audio", "text") if "audio" in feat_dict else None,
|
74 |
+
homepage=_HOMEPAGE,
|
75 |
+
license=_LICENSE,
|
76 |
+
citation=_CITATION,
|
77 |
+
)
|
78 |
+
|
79 |
+
def _split_generators(self, dl_manager):
|
80 |
+
metadata_path = dl_manager.download(_METADATA_FILENAME)
|
81 |
+
|
82 |
+
audio_paths, text_paths, metadata = [], [], []
|
83 |
+
with open(metadata_path, encoding="utf-8") as f:
|
84 |
+
for row in csv.DictReader(f):
|
85 |
+
if (
|
86 |
+
self.config.language is None
|
87 |
+
or _LANGUAGE_NAME_TO_CODE[row["Language"]] == self.config.language
|
88 |
+
):
|
89 |
+
audio_paths.append("audio/" + row["Filepath"])
|
90 |
+
text_paths.append(
|
91 |
+
"lyrics/" + os.path.splitext(row["Filepath"])[0] + ".txt"
|
92 |
+
)
|
93 |
+
metadata.append(row)
|
94 |
+
|
95 |
+
text_paths = dl_manager.download(text_paths)
|
96 |
+
audio_paths = (
|
97 |
+
dl_manager.download(audio_paths) if self.config.with_audio else None
|
98 |
+
)
|
99 |
+
|
100 |
+
return [
|
101 |
+
datasets.SplitGenerator(
|
102 |
+
name=datasets.Split.TEST,
|
103 |
+
gen_kwargs=dict(
|
104 |
+
text_paths=text_paths,
|
105 |
+
audio_paths=audio_paths,
|
106 |
+
metadata=metadata,
|
107 |
+
),
|
108 |
+
),
|
109 |
+
]
|
110 |
+
|
111 |
+
def _generate_examples(self, text_paths, audio_paths, metadata):
|
112 |
+
if audio_paths is None:
|
113 |
+
audio_paths = [None] * len(text_paths)
|
114 |
+
for text_path, audio_path, meta in zip(text_paths, audio_paths, metadata):
|
115 |
+
name = os.path.splitext(os.path.basename(text_path))[0]
|
116 |
+
with open(text_path, encoding="utf-8") as text_f:
|
117 |
+
record = {
|
118 |
+
"name": name,
|
119 |
+
"text": text_f.read(),
|
120 |
+
"language": _LANGUAGE_NAME_TO_CODE[meta["Language"]],
|
121 |
+
}
|
122 |
+
if audio_path is not None:
|
123 |
+
record["audio"] = audio_path
|
124 |
+
yield name, record
|